Research On Event Temporal Relation Classification Algorithm Based On Deep Nerual Network

Event is a common and important concept in people’s life,and can promote the development of human daily activities.The change law of the logical relationship between events is a kind of special meaningful information,especially the relationship in time.Therefore,in many possible types of relationship between events,this paper focuses on the temporal relationship between events.Time is an important aspect of knowledge representation.In natural language,time information is usually expressed as the relationship between events.Reasoning on these relationships can help to clarify the time of events,estimate the time consumed by events,and summarize the time table of a series of events.However,due to the lack of a large number of labeled training data,the application of event time relationship classification task in deep learning mainly relies on features and external knowledge base.It is easy to over fit when directly using the current network with large amount of parameters to train,and it will consume a lot of manpower and material resources if labeling.To solve this problem,this paper proposes a parallel method based on convolutional neural network(CNN)and long-term memory network(LSTM)and a hybrid network method for this task1.To solve the problem that the existing time relationship classification algorithms usually use a single depth neural network,which leads to poor classification performance,this paper uses a model combining LSTM network and CNN network for event time relationship classification.The model uses long-term and short-term memory network to extract the time sequence features of input,and uses convolutional neural network to capture the local important information of input,So as to obtain the learning ability of diversified features of various deep neural network models.It solves the problem that the existing algorithms can not extract the input sequence features and local important features at the same time,and can effectively improve the accuracy of relation classification.The model proposed in this paper is tested on matres data set,and the experimental results show that the performance of the model is improved compared with the existing models.2.For the existing models can not fully use the semantic information of the given event words,this paper proposes a hybrid network based event time relationship classification model by adding part of speech tagging to the event words in the given attribute material.In this method,part of speech tagging is trained as a joint training task to enhance the application of event description words.The experimental results show that the performance of the proposed model is better than that of the existing models in the relevant data sets.3.Build a display system of event time relationship graph.By using the above model,the extraction,classification and visual generation of event time relationship triples can be realized.